PR #9742 - simstudioai/sim

Summary

The Confluence native-query guide taught only text ~ / title ~, so assistants searching Confluence directly wrote ORed text ~ clauses. CQL returns those in no relevance order, which puts pages sharing one common word (e.g. an acronym mentioned in passing) ahead of the page about the topic.
The guide now leads with siteSearch ~, Confluence's relevance search, for topic, acronym and how-to lookups, and keeps text ~ / title ~ for exact strings. The example and the avoid line follow.
Plain (non-native) searches already fall back to siteSearch on overflow; native queries skip that fallback, so the guide is the only lever on that path.

Blind A/B on 40 real assistant-written Confluence queries: each ran as written (ORed text ~ / title ~) and rewritten as one siteSearch over the same words; a model judge compared the top 5 for the user's question in both orders, counting a win only when both orders agreed. siteSearch won 21, text ~ won 1, 18 ties (34/40 order-consistent). text ~ returned nothing in 4 cases where siteSearch returned 5. The one loss was an exact identifier lookup, which the guide still routes to text ~.
Side by side on six real topic questions, type IN (page, blogpost) AND siteSearch ~ "<words>" put the page about the topic first in five; the ORed text ~ form led with pages that only mention a term. One was mixed (a short token fuzzy-matched a longer one).
Native siteSearch queries run through the existing native-CQL path unchanged (scope, pagination, page mapping).
vitest on providers.test.ts, application.test.ts, atlassian.test.ts: 117 passed. biome check on the changed file.

Code follows project style guidelines
Self-reviewed my changes
Tests added/updated and passing (new tests pass the test-audit authoring gate)
No new warnings introduced
I confirm that I have read and agree to the terms outlined in the Contributor License Agreement (CLA)